I added the lens correction equations described here to our MATLAB code last night:
If we can figure out the "a", "b", and "c" parameters that Hugin uses (by calibrating cameras, which sounds easier than it seems to be), then we can figure out if this helps with the further away control points that weren't coming out quite right yesterday.
Without using multiple pictures or lens distortion correction, here's where we are at. (This is in our google drive document as well.) I think that we should be able to do better in the lab (these numbers might be fine for the real stream).
For square 3 in red-DSCN0094:
Error to Square (These seem nice)
Diagonal 1 0.52%
Diagonal 2 0.33%
Error lengths between control points